The use of AVA dataset

[C Li]

Hi~

   I want to add two new action classes based on this data set. But the experimental results are disappointing. I want to know whether the data set I made is not standardized, or the amount of data is not enough to cause the network to fail to extract video features. Are there any related requirements if I want to extend some new action classes. thanks a lot~